How do internships improve student major choices and early

How do internships improve student major choices and early labor
market outcomes?∗
R. Le Saout†, E.Coudin‡
August 31, 2015
Abstract
We study the role of internships in students orientation and career beginning. On-the-job experiences accumulated during schooling and before specialization, via internships, give to students some information they may
exploit to choose their majors, and to employers some insights on the skills of potential future co-workers. In
recent years, more and more French “Grandes Ecoles” students have chosen to undertake an optional full year
of internships between the two years of their Master degree. We use this selection process to assess empirically
the causal effects of internships on subsequent major choices and labor market outcomes. Internships improve
student specialization choices and fasten labor market integration. When a student undertakes several internships
in different areas, he/she is likely to refine his/her final major choices. Employers value internship experience but
less than professional experience. Employers are indeed more likely to perceive internships as a signal of ability
than as a gain in experience and human capital.
Keywords: Internships, Human capital, Returns to education, Discrete choice models.
JEL Codes: I21, I23, J24.
∗ We
would like to thank Gilles Greneche and the Conférence des Grandes Ecoles for providing us the data on the French
“Grandes Ecoles” integration surveys. We also thank Christian Belzil, Luisito Bertinelli, Victor-Emmanuel Brunel, Pierre
Courtioux, Robert Elliott, Eric Strobl, and Agustin Perez-Barahona as well as participants in the 2012 Ecole Polytechnique
Ph.d Day, 2013 Lunch Seminar at Ecole Polytechnique, 2013 Journées de micro-économie appliquée (Nice), 2014 LEMNA
Economics Seminar (Nantes), 2015 Lunch Seminar at CREST, and 2015 EALE-SOLE World Meeting (Montreal) for helpful
discussions and remarks. This paper does not reflect the views of INSEE-CREST nor of Ecole Polytechnique.
† Corresponding author. INSEE-CREST and Ecole Polytechnique, 18 bd Adolphe Pinard, Timbre L140, 75014 Paris, France.
Email adress: [email protected] Tel. 01.41.17.66.45
‡ INSEE-CREST. Email adress: [email protected]
1
1
Introduction
Internships have always represented one of the main foundations of vocational and professional formations,
e.g. apprenticeship, in contrast with traditional higher education programs. However, internships have
recently become more and more frequent in general higher education in France, far above what is required
to graduate. In particular, more and more French “Grandes Ecoles” students choose to undertake a full year
of internships between the two years of a Master degree, which is not a requirement of the Master program.
Beyond formal education, on-the-job experiences accumulated during schooling, via internships, increase the
student operational and practical skills in a professional setting. These are also privileged periods during
which students gather information about the characteristics of a given job, firm, industry, and about their
preferences for them. In parallel, employers gather information about a potential future co-worker that they
may employ at the end of his/her studies. So, internships are likely to help students to make their own
decisions about their future field of specialization and about their future career, and internships are also
likely to have an impact on wages and job conditions.
However, labor market outcomes of students who undertook a full year of internships do not apparently
differ from those of students who did not. In forty-five engineering schools for which this year of internship is
common but not compulsory, we found no effect on satisfaction at work and only a small increase of wage with
a 3% premium, which is far below the returns to the first years of experience in such schools.1 Only 2% more
students who did a full year of internships find a job before graduating (50% versus 48%). These apparently
low benefits are unlikely to justify to postpone labor market entry of one later with a financial loss. Hence,
these aggregated comparisons do not enable us to understand the huge increase in one-year internships. Using
detailed data on labor market outcomes of new graduates of one “Grande Ecole” matched with their schooling
attainment during the master program, we study the effect of internships undertaken during general higher
education studies, on student choices about their specialization, and on their employability and wages after
graduation. While many political debates were held on internships in France, little is known about their
potential effects. In the French engineering schools, students can make the choice to undertake an optional
full year of internship between the two years of the Master. This possibility teaches us about how this first
work experience is perceived by firms, as a signal of student quality or as a gain of human capital, which is
a sensitive issue in economics.
Internships were largely studied in sciences of education and psychology. Taylor (1988) underlines that
internships improve employment opportunities, and in a more partial way, knowledge of interest and work
values. Some management studies (Gerken et al. (2010), Narayanan et al. (2010) and Gault et al. (2000))
focus on internships in business schools, but are mainly based on qualitative and descriptive analysis. Firms,
universities and students can gain from effective internships. For firms, internships may be help for special
projects and a signal on future employees. Universities get better connected to the corporate world and may
improve their reputation. Students may reduce job search duration and increase their knowledge of interest,
work values and potentially wages. Gault et al. (2000) survey 223 students graduating during 1994-1996 and
find positive effects on wages (9% higher) and satisfaction at work, and negative effect on job search duration
(less of more than 2 months). In economics, Saniter and Sandler (2014) find a positive effect on wages in the
German labor market, using IV analysis with the introduction of mandatory internships in some programs.
1 These
figures come from surveys "integration of graduates" described below.
2
Using an experiment with CVs including internships as one feature, Nunley et al. (2014) find too a positive
effect on the interview rate, which may be explained by signaling. However most of the literature on youth
training in the workplace focuses on apprenticeship, that is, fully designed programs alternating workplace
based training and formal education periods,2 see Wolter-Ryan (2011) for a complete recent review and the
seminal work of Becker (1962).3 The main concern of these studies is to account for the endogenous selection
of individuals between apprenticeship and full-time formal vocational education. Our approach is close, as
we consider the optional full year and the major choice as an endogenous selection process too.
A closely related literature concerns the determinants of college majors and associated returns (see Altonji
et al. (2012) for a complete review and a general but mainly theoretical approach of sequential-decision
schooling models). Education decisions are made sequentially and outcomes are uncertain, see Altonji (1993),
Heckman et al. (2003). Preferences, beliefs, abilities, expected earnings and non-pecuniary employment
conditions revealed along the schooling process, are expected to affect student choices at the next step of the
schooling process, and especially when they choose their specialization majors. Arcidiacono (2004) analyzes
the role of ability in major choices, distinguishing verbal and maths abilities. Stinebrickner (2003) shows
that students change majors after observing their grade performance in the first years. For France, Beffy
et al. (2012) find very low, though significant, elasticity of college major choice to expected earnings and
conclude that non-pecuniary factors are the key determinants. For the U.S., Arcidiacono et al. (2012) show
that uncertainty about wage expectations generate inefficient decisions. Optional internships may provide
the occasion for students to revise their beliefs about their preferences, their occupation-specific abilities,
their expected earnings and some non-pecuniary employment conditions. They consequently help to reduce
the uncertainty of related decisions.
We analyze the effect of a full year of internships4 in a French Grande Ecole on student subsequent choice
of specialization and labor market outcomes after graduation, i.e. job search duration, wages, satisfaction at
work. Our data come from an original survey about the labor market integration of graduates matched with
the students’ attainments during their schooling. This enables us to account for students abilities. We consider
that schooling decisions are sequential. The decision of undertaking an optional full year of internships is
indeed done before the major choice one. During their first years of schooling (or other experience), students
receive information about their abilities and preferences. At the end of the first year of the Master degree,
they decide whether or not they would do an optional year of internships between the two years of their
Master degree. At the beginning of the second year, they choose a major after having or not done a year of
internships the year before. Graduates finally enter the labor market or decide to go on their studies (Ph.D
2 Wolter-Ryan (2011) define “apprenticeship as programs that comprise both work-based training and formal education, in
most countries at upper-secondary level, and lead to a qualification in an intermediate skill, not just to semiskilled labor.”
3 The questions addressed include where to develop specific-occupation skills in the workplace or at school, the firm behavior
in entering and financing an apprenticeship system, the outcomes for apprentices. Answers greatly depend on the countrylevel institutional context. Some other papers compare secondary education apprenticeship to full-time formal vocational high
school in terms of labor market outcomes, especially risk of unemployment. In France, Bonnal el al. (2003) show a positive
impact of apprenticeship versus full-formal vocational high school on getting rapidly a first job, and Alet and Bonnal (2011),
a positive impact on subsequent schooling outcomes. Studies are more numerous in German speaking countries, in which the
system of apprenticeship is widespread. Developing a full structural approach, Adda et al. (2010) find an increase in wages
and an accelerated experience gain profile in Germany. Exploiting small firms closures and consequent exogenous variations in
apprenticeship durations, Fersterer el al. (2008) find a 5% pay increase of apprenticeship. A parallel literature focuses on the
advantages of developing some occupation-specific skills, through some vocational courses, in general high school, to increase
students’ cognition and motivation, see, e.g., Altonji (1995), Mane (1999) and Bishop and Mane (2004).
4 In the article, the terms “full year of internships” refer to one or more internships for a duration of one year and three
months (“long internship”), which replace a mandatory 3 months internship.
3
or other).
The sequential schooling decisions framework highlights several potential causal effects of internships,
which we test using reduced-forms models. Students who undertook a full year of internships between the
2nd and the 3rd year of their program may have higher wages or reduce their job search duration at their labor
market entry for several reasons.1) It may be a signal to employers. With similar characteristics and in the
absence of other available information, employers are likely to choose graduates who completed internships
in their firms since they have information on student individual ability, which is usually unobserved. 2) The
experience accumulated as an intern may be valued by employers, which can be compared to the return to
the first years of experience. 3) Doing a full year of internships may improve opportunity of careers (a first
job effect), the professional network and the returns to experience. Hence, we can compare the return to
experience with and without doing a full year of internship. 4) Students may have improved their orientation,
and may have chosen their major specialization more accurately according to their preferences and abilities.
The matching quality between graduates and first jobs may be improved. Students who did a full year of
internships are likely to be more satisfied of their job. 5) Students may have delayed their entry into the
labor market in times of economic slowdowns. This effect is nevertheless expected to be weak (Gaini et al.,
2013).
Empirical results indicate that the gross wage return of a full year of internships values is smaller than
75% of the return of one of the first years of work experience. Hence, this is lower than the return of a real
year of professional experience. The effect is mainly determined by the revelation of abilities. Internships
are perceived as a signal of ability by employers more than as a return to experience and a gain in human
capital. The econometric analysis takes into account the ability bias and the endogeneity of the choice of
making a full year of internship. With IV analysis, the wage return of a full year of internships appears
higher than a real year of professional experience, but this result has only a local interpretation. This year of
internship improves the ability to find a job faster. When a student undertakes several internships in different
areas, he/she is likely to refine his/her final major choices. The effects on job satisfaction remain ambiguous,
positive but small and not significant. Public policies implemented, which ban unpaid internships or more
than six months, appear consistent with our results. Compensation for internships of one third of potential
wage output seems necessary to financially equalize the loss. Perform a full year internship should help to
refine these major choices. It therefore seems logical to ban internships more than 6 months, which can be
considered as a real job and not as an internship.
The layout of the paper is as follows. Section 2 contains contextual elements about the French higher
education system with Grandes Ecoles and universities and the place of internships therein. Section 2 also
describes the schooling program of the Grande Ecole studied. Section 3 presents the data used. Descriptive
statistics are reported in section 4. Section 5 presents the reduced-form estimation strategy and results.
Section 6 concludes.
4
2
Institutional context
2.1
Overview of French higher education system
Universities and Grandes Ecoles. (cf. figure 1) In France, two types of higher education institutions can
deliver graduate degrees: universities, which are large state-financed structures, and "Grandes Ecoles", which
may be state-financed or private and are of smaller size. The "Grandes Ecoles" contain "engineering schools"
and "business schools". Their particularity is that they are generally selective contrary to universities. To
get admitted, students prepare for competitive examination (“concours”) during two years. This competitive
entry and the ranks of admitted students contribute a lot to the reputation of the Grande Ecole and to the
delivered degrees. French Grandes Ecoles students represent about 15% of students in higher education in
2010 (about 8% in 1980). Some of the "Grandes Ecoles" are quite generalists, others are more specialized.
Their schooling programs usually last three years, leading to a master degree.5 Even though there is a lot of
variability between Grandes Ecoles (and universities), labor market outcomes of Grandes Ecoles students are
on average better than those of university students. In 2010, six months after graduation, 84% of students
from "Grandes Ecoles" are employed and earn on average 35.800 euros (bonus included). The selection at
the entrance is one argument in favor of these higher labor market outcomes, the fact that Grandes Ecoles
programs contain on-the-job training periods is another. A Grande Ecole curriculum is usually composed of
periods of formal education and compulsory periods of internships. The load of each, and the sequence in
which they intervene, depend on the program. Even though curriculum contains some work-place experience
periods, Grandes Ecole schooling programs cannot be considered as vocational training, nor apprenticeship.
The latter correspond most often to certifications below the bachelor degree.
6
Internships in the French higher education system. Internships are compulsory part of many training programs, but their duration and content really depend on the program, and the type of program. Internships
are constitutive and widespread in Grandes Ecoles programs. Programs in these schools are labeled by external commissions, which check and insure that the program fits certain requirements, including a certain
number of weeks of internships. Those labels are very important for schools as they entail the real recognition of a certain level of education and degree equivalences. So all schools do their best in order to get
their programs to satisfy those requirements. Most Grandes Ecoles contain an "internships office", which
is composed of administrative staff who receive, find, check, and screen internship offers sent by firms and
transmit them to the students. In 2010, while almost all students from "Grandes Ecoles" undertook at least
one internship (there is usually a compulsory 6-months internship at the end of the last year of the program),
university students were only 61% to undertake an internship in the last year of the Master. 97% of students
in "Grandes Ecoles" were paid for their last year internships versus 76% for university students of equivalent
education level. 57% earned more than 600 euros a month versus 29% for university ones.
In the recent years, a policy debate in France has accompanied a change in regulation about internship
minimum remuneration and internship duration. The drawbacks of unpaid internships were discussed. First,
as stated by Richmond (2010) for the UK, students with lower financial resources would have fewer opportunities to undertake unpaid internships than their counterparts, which may in turn deteriorate their entry
5 The
grade of Engineer is in France equivalent to a master degree.
apprenticeship system has been greatly developed in France in recent years, although it remains low in the higher
education: apprentices accounted for 0.9% of students in higher education in 1995, 2.4% in 2000 and 4.8% in 2010. The majority
of these apprentices are under the bachelor degree level (90% in 1995, 75% in 2010).
6 The
5
conditions on the labor market. Second, unpaid internships may also be viewed as unfair competition with
older workers for whom minimum wage regulation applies. The Law for equality of opportunity of March
2006, implemented in February 2008, forbids unpaid internships longer than three months. This regulation
has been extended in November 2009 to forbid unpaid internships lasting longer than two months. Internships
have to be paid at least 400 euros per month, which represents a third of the minimum wage. The Cherpion
Law (2011) and the Fioraso Law (2014) prohibit (paid) internships, which last more than six months and
strengthen intern rights but these laws are not yet implemented. In contrast, in the U.K. or in the U.S., there
is no legal obligation for firms to pay an internship that lasts less than one year if the latter is required as
part of study curriculum. However, if interns satisfy the legal definition of ‘workers’ in the United Kingdom
(i.e. they do work for which other employees are paid) or some equivalent requirements defined from the Fair
Labor Standards Act in the U.S., they should normally be paid at least the minimum wage.
2.2
French Grande Ecole schooling program
We now describe more precisely the Grande Ecole (ENSAE-ParisTech, Paris Graduate School on Economics,
Statistics and Finance), on which the data rely, and the schooling program. The school is a relatively small
Grande Ecole. Around 150 students graduate each year. Students who succeed the examination entrance
enter the first year of the curriculum. This first year differs slightly between students upon their previous
studies. Those who have studied economics before are taught more maths and those who have studied
maths before, more economics. Some other students, selected on records, directly enter the second year
of the program. The second year of the program is composed of compulsory common general courses in
Economics, Statistics, and Humanities and students choose a minor in Economics or Applied Mathematics.
The minor choice is made by the student upon his/her preferences and expected abilities. There are no
quota in registration in a minor, neither restrictions based upon previous specialization. At the end of the
2nd year, students choose either to undertake a compulsory 3-month internship and to enter the 3rd year of
the program, or to undertake internships for a complete year and three months. If so, they come back one
year later in the 3rd year of the program, and graduate one year later. There are seven majors in 3rd year,
which, due to the small number of students per major, we will gather in two main fields, Finance/Actuarial
Sciences and Economics/Statistics/Social sciences. Those are equitably distributed with 54% of students in
the Finance-oriented majors and 46% in the other ones. Students choose the 3rd-year major at the beginning
of the 3rd year. Again, there is no quota, whatever her previous curriculum is, a student enters the major
he chooses.
3
Data
We use five "integration of ENSAE graduates" surveys (2009 to 2013), which we match with the administrative
data on student schooling attainments.7 The survey " integration of graduates" is organized every year in all
French "Grandes Ecoles", so approximately 200 schools (including both "engineering schools" and "business
schools") and 40,000 students. The survey focuses on the labor market integration and job conditions at the
7 The survey has been declared to the French “Commission nationale de l’informatique et des libertés” (declaration number
1604776) with the precision that the survey data can be matched with data coming from the administrative education software
(“pamplemousse”). All data are anonymous.
6
date of the survey, i.e. wage and job satisfaction. The survey questions exhaustively the two (for surveys
2009 and 2010) or three (for surveys 2011, 2012, and 2013) promotions who graduated during the year, the
year before or two years before. A student is then interviewed 6 months, 18 months, and 30 months after
graduation. Each year, the questions are identical and use the same definitions to allow time comparisons.
The survey data are matched to schooling attainment records that come from the education management
software used. These administrative data also contain student 3rd-year-major specialization choices, 2ndyear-minor choices; and some basic information on student previous studies and on their social-demographic
characteristics: parental background and location, social scholarships, age, gender, etc..
The sample contains students admitted in the first year or in the second year of the program as they are
the one for whom the full year of internship option is opened. We exclude all students for whom the full
year of internships option was not opened, i.e. students who joined the program directly for the third year
(around 23% of a graduating promotion); civil servants (14%); students who were registered in double degree
programs in convention with other institutions (around 12%); the very few students who repeated the first
or the second grade (less than 4%).
Finally, the dataset is composed of 452 students who graduated between 2007 and 2012. 39% are women.
65% were admitted in the first year, 35% directly in the second year (i.e. the first year of a Master degree).
The overall response rate is around 71%, lowering with the time after graduation: 82% for the more recent
graduating promotions, versus 65% for the others. This overall response rate is between 74% and 78% for
surveys 2009 to 2012, but only 57% in 2013. The comparison between the survey response rate and the
numbers of students per 3rd-year major confirms that the sample of respondents is representative. Student
who majored in Economics/Statistics/Social Sciences respond more frequently than those who majored in
Finance (81% versus 64%). We construct weights to correct for non-response, and estimate both weighted
and non-weighted regressions.8 Results from weighted and non-weighted regressions are close. Partial nonresponse is very low, and individual responses between surveys are quite consistent.9 Job satisfaction was
not surveyed in 2009.
Variables of interest. Using administrative schooling management software data, we construct a dummy
variable indicating whether a student undertook or not a full year of internships. We also construct 3rd-year
major variables. In contrast, labor market outcome variables come from survey data.
As wage variable, we use the annual total of wages including bonuses in real terms, deflated using price
consumption index. On average, 62% of people report bonuses, which consists of 12% of their annual
remuneration. Bonuses increase with experience. Further, we construct proxies for job satisfaction ranging
from 1 (worst) to 5 (best): satisfaction regarding relations with colleagues (working environment), level of
8 We construct unequal weights, obtained as inverses of probability response estimates of a Logit, and stratified weights, based
on the year of graduation, the major and the duration the survey year and the year of graduation.
9 Partial non-response is equal for example to 6% for duration of the first job search and to 10% for wages. Students are
surveyed about their current job. However, there are additional questions for those for whom this is not the first job. So,
depending on the fact that he/she changes or not of job, we either have repetitive information on the same job or information
up to 4 different jobs he/she held. 78 students (17%) did not give any information, 157 (35%) one response, 83 (18%) two
responses, 99 (22%) three responses and 35 (8%) four responses. So some of them answered several times at the same questions.
We analyzed the consistency of responses and memory bias. Concerning the first job search duration, answers between surveys
are consistent. They differ in less than 10% of the cases and in such cases for minimal differences: answering “between 2 and 4
months” instead of “between 4 and 6 months”, for example.
7
autonomy in work, working conditions, and salary level. We consider a worker satisfied if he chooses the
response item 4 or 5 (the scale ranges from 1 to 5)10 .
We have included job characteristics in the analysis, such as working abroad (without distinguishing
between countries), working in the public sector, or working with a fixed-term contract. We do not include
the firm size nor business industries because the data upon obtained in the survey were of poor quality. The
firm size is often misinformed because of confusion between responding at the firm or the local unit level.
However, we think that this omitted variable has little impact on the results since the majority of graduates
work in very big firms (banks and assurances). 18% of graduates finally work in another industry than the one
they majored in (a student majoring in economics who works in the banking sector for example). However,
we cannot control precisely for the industry as the survey do not manage to differentiate close professional
industries.
Moreover, we do not observe how employers define their wage proposal. The bargaining power of graduates
may be reduced if employers use fixed pay scales based on the name of the university or the “Grande Ecole”
followed, independently of whether the graduate did or not a full year of internships. This may explain a
lower effect of internships. We do not know whether the graduate works in the firm where he/she did an
internship because the name of the firm is misinformed. We cannot therefore exclude the hypothesis that the
effect of internships differs regarding to the fact that graduates work in the same company as the one they
did an internship in or not.
Because of majors, students do not take the same courses during the 3rd year of the program. Hence,
their scores are not comparable. In order to construct proxies for abilities, we rather rely on 2nd year
achievements. We compute inverse ranking scores based only on the schooling attainments upon common
courses. The best student is ranked 100 and the last one 0. We construct a global ranking based on the
average score during the year (all courses including group projects, and after the catch-up examinations),
and specific scores based on written individual evaluation in mathematics (statistics and time series), and
economics (micro- and macro-economics). Courses to be included in the specific rankings have been selected
using a principal component analysis.
4
4.1
Descriptive statistics
The choice of doing an optional full year of internships
Figure 2 reports the increasing share of students choosing to undertake an optional full year of internships
between the 2nd and the 3rd year of the curriculum. 23% of those who graduated in 2007 used this option
and 62% of those who graduated in 2011. 52% of students admitted in the first year completed a full year
of internships (32% in 2007 and 76% in 2011) whereas only 17% of those admitted in the second year did
so (almost none before 2010 and 38% in 2011). Furthermore, 56% of students who opted for the full year
10 Responses 1 or 2 are rare. There are 3% of responses 1 for wages, and 1% for the non-monetary variables. There are 9%
of responses 2 for wages, 6% for autonomy and the working conditions, and 2% for the working environment. The response
item 4 seems the more stable, with 36% of response for wage, autonomy and the working conditions, and 31% for the working
environment. The highest satisfaction (5) is obtained for the working environment (57%), then the working conditions (44%),
autonomy (38%) and finally wages (28%).
8
of internships option undertook two internships, compared to 37% just one and 8% three. Internships in
Finance are more frequent. Generally, internship topics are related to minor and major choices.
4.2
Minor, major choices and schooling attainments
Table 1 compares student rankings and major choices depending on the fact that students did or not a full year
of internships. As noted before, 54% of the students major in Finance and 46% in Economics/Statistics/Social
Sciences. More students who did a full year of internships chose to major in Finance during the 3rd year of
the program (61% of cases vs. 49%) whereas there are no apparent relation between 2nd-year minors and
the choice of undertaking or not a full year of internships.
Students who chose to undertake a full year of internships seem to have lower 2nd year schooling attainments on average than others. Their average inverse ranking is 46.1 versus 53.0 for those who did not. Note
that we report an inverse ranking to make rankings comparable to GPA. This holds regardless of the 2nd-year
minors (5.7 ranks for mathematics; 9.6 ranks for economics) and of the compulsory courses considered. Students choose their specialization in line with their skills. Those who minor in economics have better ranking
in economics courses than in mathematics courses (54.4 versus 41.5), and vice versa for those who minor in
mathematics (56.9 average ranking for mathematics courses versus 51.9 for economic courses).
4.3
Early labor market outcomes
Table 2 reports average labor market outcomes of graduates according to whether or not they undertook a
full-year of internships. This option apparently fasten labor market integration: six months after graduating,
only 2% of graduates who did a full year of internships were looking for a job compared to 9% of those
who did not. The employment rate is higher 73% including volunteering versus 55%. Graduates usually end
their studies more rapidly: only 11% of them continued their studies after graduation against 19% for those
who did not the full year of internships. On average over the three promotions interviewed in each survey,
students who did a full year of internships earn 13% more than those who did not (53.300 euros bonuses
included versus 47.100 euros). Wages rose sharply in the early years of the career. They increase of 13% on
average between the first and second year, and of 7% between the second and third year after graduation.
This evolution differs nevertheless according to whether or not graduates undertook a full-year of internships.
The increases are 28% (between the first and second year) and null (between the second year and the third
year) for graduates who did a full year of internships while they are 7% and 13% for others. This may be due
to a later entry into professional life. Students majoring in Finance earn on average 12% more. There is no
correlation (or only a small one) between a full year of internships and the wage differential by specialization
fields six months after graduation. These figures suggest that the on-the-job experience acquired during
the year of internships is valued by the employers. Concerning satisfaction at work, the results are less
obvious. Internships seem to be associated to a better satisfaction (for all components) but the difference
is small. There is more satisfaction for non-monetary components (72% for autonomy, 80% for working
conditions, and 88% for working environment) than for wages (63%). However, when defining satisfaction
for non-monetary components with a score higher or equal than 4 for each component, the overall rate of
non-monetary satisfaction is only 60%. If we compare satisfaction between responses of students with and
without a full year internship, we do not find a significant effect at 10%. The difference is however around
9
10% for autonomy in work, and for overall non-monetary satisfaction (i.e. for 100 workers, 10 would answer
4 or 5 instead of 3 for example). For wages, working environment, and working conditions, differences are
less than 3%.
Table 2 reports the distribution of the first job search duration. 63% of the graduates who did a full
year of internships found a job before graduating compared to 53% of those who did not. They are also less
likely to have a search period lasting more than 6 months, 4% versus 10%. 28% of the students declare that
internships - including full-year, part-time and other compulsory ones - helped them to find a job. Full-year
of internships are also associated to longer job periods after graduation. The first job - when ended - was
also on average longer - of 3 months- for the graduates who did a full year of internships. These figures
suggest a better matching between the graduates who chose to undertake a full year of internships and the
employers occurred. 18% of students studying in a major work finally in an other employment sector (a
student majoring in economics who works in the banking sector for example). As this figure is low and the
distinction between close professional sectors difficult, we assume the major choice is associated to a choice
of employment sector.
5
Internships, major choice, and early labor market outcomes:
results
5.1
Estimation strategy
Our goal is to test the potential causal effects of internships. There are two endogenous variables in the
analysis, undertaking or not a full year of internships and majoring in Finance. We consider indeed that
schooling decisions are sequential. The decision of undertaking an optional full year of internships is done
before the major choice one, depending both on future expectations on wages and non-monetary benefits. As
the labor market can be considered as an absorbing state (you cannot return to school after entering in the
labor market), the effects of undertaking an optional complete year of internships on labor market outcomes
after graduation can then be tested using a Mincer type model for wages (and close models for duration of
job search and satisfaction at work) and controlling for the endogenous selection with IV analysis and 2SLS.
The subsequent schooling decision of the major choice is estimated with a classical probit model. To be valid,
Rust (1987) assumptions are the conditional independence (the unobserved factors at t have no effects one
period later if you control by the decision variable at t + 1) and that unobserved heterogeneity is a random
effect. It means that there is no omitted variable and no serial correlation of unobserved preferences. In
order to have weaker assumptions, we estimate a biprobit model of the both subsequent schooling decisions
(major and full year internship).
5.2
Internships and major choice
We examine the determinants of the 3rd-year major choice and the decision to do a full year of internship.
We assume that individuals choose their majors to maximize a value function Vm , which is the sum of the
non-monetary benefits and the present value of expected earnings, m? = argmaxVm . These both terms
m
are unknown by the student and approximated when learning of his own ability and preferences. The non-
10
monetary benefits are assumed to be a linear function of observable individual covariates (such as ability for
a special topic, gender...). The present value of expected earnings may depend of the economic conditions.
There are 7 different majors in 3rd year of the program but due to the small number of students per
major, we group them in those related to Finance/Acturial Sciences and those related to Economics/Social
Sciences/Statistics. We explain the 3rd year major by students schooling attainments in 2nd year (inverse
ranking i.e. 100 for the best student, 0 for the worst), minor choices in 2nd year, a dummy identifying those
who chose to undertake the optional full year of internships, and the topics of the internships undertaken.
Demographic characteristics (sex, place of birth) are used as control variables. The constant of the probit
model for majoring in Finance can be interpreted as a wage differential, and so on an estimation of expected
earnings. We add the year of graduation, which can measure the economic conditions11 . The major choice and
the decision to do a full year of internships may be explained by identical unobserved factors. Furthermore,
there may be serial correlation between unobserved preferences of these two endogenous variables. We so
estimate a biprobit model of these both subsequent schooling decisions. The student admission type and
his/her age in the first year of master are used as instruments to explain the decision to do a full year of
internships. The choice of instruments is discussed in the IV analysis of the wages regressions.
Table 3 reports the average marginal effects of a probit model for a major choice in Finance. Second year
minor choice is naturally correlated with the 3rd year major one. Students who chose a minor in applied
maths are more likely to major in Finance in 3rd year with an average marginal effect of 29-53%. The
minor choice is a first expression of students’ preferences and expected abilities. As seen previously in the
descriptive statistics, students who undertook a full year of internships are more likely to major in Finance
with an average marginal effect of 14%. However, when a student undertakes several internships in different
areas, he/she is likely to refine his/her final major choices. Doing several internships both in economics and
finance reduces the probability of majoring in Finance around 13%. 2nd year general ranking is positively
correlated with majoring in Finance, but the results are not significantly different from 0. In contrast, the
mathematics ability is highly correlated with majoring in Finance. The decision to do a full year of internship
is not correlated with the second year minor choice (column 6). The admission in 2nd year, the age in 2nd
year, and the rank have a significant and negative effect. Increase its ranking of one decile decreases the
probability of 2%. Concerning economic conditions, we observe that graduating after 2010 increases the
decision of doing a full year of internships. We do not observe such cyclical choices (or only weakly) for the
major choices. When the decisions are estimated simultaneously with a biprobit (column 5 and 6), the effect
of the full year of internship on the major choice is reduced despite the fact that the correlation between the
error terms of the two equations is not statistically significant. This could be because the causal effect of
internships in students orientation is mainly observed for students with internships in different areas.
5.3
Labor market outcomes: duration and wages
The choice of the number of years of schooling has already been done by the student at the entry of the
"Grande Ecole" (more than 5 years depending on whether the student goes on with a Ph.D or not). The
11 To help them to make these decisions, the administrative staff informs students on the labor market outcomes of former
graduates by communicating about the results of the last Integration of Graduates survey. But results are published two years
after graduation and are not known over a long period (in particular before 2007).
11
failure rate is very low in French “Grandes Ecoles” and the degree is obtained 5 years after high school
graduation. The wage of a graduate on the labor market is defined with a Mincer type analysis12
log (Wmt ) = α0mt + α1mt · d + α2m Am + α3m · d · Am + α4m expt + α5m · d · expt + Xt β + εmt
with t, the year, m, the third year major, log (Wmt ) the log of total wages - bonuses included, d, a dummy
equal to one if the student did a full year of internships, A, the student ability, which is major specific (a
student may choose a major where his/her ability is higher) and can be split into a general and a specific
ability, exp the experience in the labor market (a squared term is not necessary for the first years), and X a
vector of other characteristics (job characteristics (working abroad, second job since graduation, whether it is
in the public sector, whether the job contract is fixed-term or long-term), and student characteristics (gender,
nationality, year of graduation).
Experience (in years 0, 1 or 2) accounts for the sum of the duration of
each job reported in the survey. The year of graduation accounts for effects of economic conditions. Students
ability is measured by their ranking in the second year (the courses are partly common unlike the third year),
or their ranking on specific courses (economics and mathematics), which has been defined using a principal
component analysis. As robustness check, we also used a potential experience variable.13 X is assumed to
have the same returns for each major and whether or not the student has done a full year of internships. The
other variables are major- and internship- specific. The constant term measures the structural differences of
wage returns between majors and the gain associated with a full year of internships (to be compared to an
additional year of experience).
This reduced-form model entails some predictions that can be tested.
• Signaling to employers: with similar characteristics and in the absence of other available information,
firms are likely to choose graduates who completed internships. The ability of the student A is largely
unknown by the firm. Doing an internship so reveals a part of the student ability to the employer, with
α2m = 0 and α3m > 0 in an extreme case where the ability is completely unknown without internship
and revealed with internship;
• A cost-benefit analysis between the opportunity cost of entering the labor market one year later, and
the wage gain associated with the experience accumulated as an intern: there would be a wage return
to a full year of internships. We can so compare the return to experience α1 and the return to a full
year of internship α4 .
12 A huge literature has focused on the returns to education, with estimates based on the traditional Mincer equation, which
explains wages by years of schooling, experience and other control factors. Due to the potential endogeneity of the schooling
decision (as the student ability is not observed), IV estimators have been proposed with instruments mainly based on compulsory
schooling laws. This approach has been debated (see Björklund and Kjellström (2002), Dickson and Harmon (2011) for a brief
review). Heckman et al. (2003) analyze the assumptions of the classical Mincer model and conclude that derived returns to
schooling estimates are not valid. Indeed, this static model neglects the major determinants of schooling decision such as the
uncertainty about the future wages, the cost of schooling, or the variation of these returns over the life cycle. Furthermore,
the use of IV estimators to address the problem of the omitted ability bias is not powerful, and quite different results occur
depending on which instruments are used and which assumptions are made (Heckman and Urzua, 2010). Blundell et al. (2005)
compare several methods linked with the evaluation literature: matching, IV, and control function approach. They emphasize
the importance of the detailed test score and the family background differences to explain (heterogeneous) schooling returns.
Non-monetary benefits of education are generally not taken into account, despite the fact that health and well-being can be
improved with a higher education (Heveman and Wolfe, 1984).
13 The Pearson correlation between the two experience variables is very high, 90%.
12
• A better return of experience: doing a full year of internships may improve opportunity of careers (a
first job effect), the professional network and the returns to experience. The returns to experience also
depend on first job opportunities, which may differ depending on internships done and majors chosen.
Hence, we can compare the return to experience α5 with and without doing a full year of internship.
It is also possible that students wait for good economic conditions. They could delay their entry into the
labor market if they think the economy is not favorable. We so add the year of graduation as a control
variable for the duration of job search after graduating.
A similar analysis can be done for the duration of job search after graduating. Table 4 reports the
average marginal effects of a probit model for finding a job before graduation (column 1) and shows that a
full year internship reduces this duration but not by revealing the ability of students (interaction effects are
not included but are close to 0 and non significant). With an interval regression (column 5), this reduction
of job search is estimated less than 2 months. In column 2, IV analysis with a linear probability model shows
that the effect may be underestimated but becomes not significant (the choice of instruments are discussed
thereafter). The economic conditions (measured by the year of graduation) have effects on duration of job
search.
Table 5 reports wages determinants. In column 1, we compare the effect of a full year of internships to the
returns of the first years of experience. In column 2, we add education characteristics (3rd year major chosen,
inverse ranking). In such analyzes, there may be an omitted ability bias. Students have different skills, often
unobserved. We add student achievements in the program. We use the inverse ranking of student in 2nd
year i.e. 100 for the best student, 0 for the worst, as proxy for ability. In column 3, we use a LAD estimator
(median regression) to take into account extreme wages. In column 4, we separate the overall ability in two
specific skills, mathematics and economics. In column 8, we cross the full year internship dummy with each
year of professional experience (0, 1 or 2) in order to check if the effect could disappear with experience. A full
year of internships has a positive and significant effect around 8% on wages, which is reduced by half when
we control for extreme wages. A rank increase of general ability by one decile corresponds to a pay increase
of 0.5% more, due to the economics skills (as the mathematics skills have an unexpected negative value) with
a 1.3% premia. Experience has a positive and significant effect higher than 10% on wages, reduced around
9% when we control for extreme wages. Consequently, with a static vision, one may infer that the wage
cost-benefit trade-off of undertaking a full year of internships is negative; it would be rewarded less than the
year of labor market experience it replaces, smaller than 75% and even 43% when controlling for extreme
wages. The students’ enthusiasm for it would remain a puzzle. The effect of a full year internship does not
disappear with experience (column 8). Choosing to major in Finance is associated to 9-10% premia. Finally,
over 2007-2012, the year of graduation has no effect and has not been included.
The effect of a full year of internship or to major in Finance could differ following ability of students.
Furthermore, returns to experience could not be the same. We test those assumptions by including interaction
effects in our analysis. Results are reported in table 5. Columns 5, 6 and 7 report the OLS estimates. Column
1 adds interaction effects between education characteristics (full year of internship), experience and a third
year major in Finance. In column 5, the effect of a full year internship would be the same for students
majoring in Finance and in other major. Choosing to major in Finance is associated to a non-significant
premia. Concerning returns to experience, we note that returns are higher for students majoring in Finance
13
(around 6%), and small and non significant for a full year of internship. Concerning returns to ability, a
full year of internship would have a positive effect to reveal abilities. In column 6, the effect of general
ability is null for students who have not made a full year internship but highly positive and significant for
others. A rank increase of general ability by one decile corresponds to a pay increase of 2% more. In column
8, concerning specific abilities, we note that it reveals mainly the mathematics abilities (there could be an
adverse selection effect for students who have not made a full year internship).
In our sequential decision framework, the internship take-up depends on future expectations on wages and
non-monetary benefits. Those gains can be explained by several factors, such as heterogeneous preferences,
beliefs, risk aversion, or individual abilities. We control our econometrics analysis with schooling attainments,
which can be interpreted as a measure of individual ability. However, there may be an omitted part of
individual ability, and there may also be some error measurements. Ability for theoretical courses does
not say anything about communication skills for example. If the omitted ability is positively correlated
with the schooling attainments (i.e. students with better schooling attainments have also a better ability to
communicate) and as students who take-up a full year internship have lower schooling results, there would be
a downward bias. However as the sign of this correlation is unknown, the direction of the bias is uncertain.
Furthermore, if students who did a full year internship were those who have preferences for professional
settings and lower preferences for non-monetary benefits, it would interact with monetary and non-monetary
benefits. Those students may engage in jobs with better wages and lower working conditions, which may
create an upward bias in the wage equation. All in all, the bias may exist but the direction of the bias is
largely uncertain.
There are two endogenous variables in the analysis, undertaking or not a full year of internships and
majoring in Finance. They depend both on future expectations on wages and non monetary benefits, which
can be explained by several factors, such as heterogeneous preferences, beliefs, risk aversion, or individual
abilities. In an attempt to control endogeneity for the full year of internships and a major in finance, we
conduct an IV analysis. There are two main identifying assumptions behind the IV equations.
The first one is that the instruments do not have to be correlated with the dependent variable (i.e. the
exogeneity condition), such as wage, the probability to find a job before graduating or satisfaction at work.
Our set of instruments is the student age at the end of the 2nd year of the program, the year of admission, a
year of graduation higher than 2010 and the minor choice in 2nd year. Our assumption is these variables are
not correlated with wage, the probability to find a job before graduating (except for the year of graduation)
or satisfaction at work. In France, each “Grande Ecole” is similar to a brand valuated by firms, for which
a shadow “ranking” is obtained according to the difficulties to be admitted. We may assume that, as this
school is highly selective, that employability and wages do not depend of past experience such as being
graduate from an other university for example. The student age and the year of admission would be thus
valid instruments for undertaking a full year of internship. Similarly one may think that firms consider
mainly major choices and the “Grande Ecole” standing to fix wages or to hire an employee. The minor choice
would be thus uncorrelated with wages and be a valid instrument to explain the major choice. Finally we
may think that the economic situation primarily affects employability but not wages which are considered
rigid even in a recession. A year of graduation higher than 2010 would be then a valid instrument for wages
but not for the probability to find a job before graduating. These assumptions are questionable. If violated,
there will be no endogeneity correction with an IV analysis. However, if there are as many instruments
14
as endogenous variables, this exogeneity condition cannot be tested. If there are more instruments than
endogenous variables, a Sargan-Hansen test of overidentifying restrictions can be performed, whose the joint
null hypothesis is that the instruments are consistent instruments (i.e. the choice of instruments is not
invalidated).
The second condition is that the instruments have to explain the endogenous variables (i.e. the rank
condition), which is tested in the first stages regressions of 2SLS. The student age at the end of the 2nd year
of the program or the year of admission is likely to be negatively correlated with the full year of internships.
It will be more expensive for older students to postpone their entry into the labor market. Due to the
implementation of laws banning unpaid internships starting in 2008, a year of graduation higher than 2010
(i.e. two years after the law) can be a possible instrument too. The minor choice in 2nd year is a natural
instrument for the choice of major. All F-stat are higher than 10. As there are two endogenous variables,
a global F-stat is computed who gives the same conclusions. However, we exclude potential instruments
because they were found to be weak instruments, such as having a social scholarship or the value index
for French banking and insurance equities. With weak instruments, results can be indeed misleading. Due
do financial constraints and the fact that undertaking a full year of internships implies postponing labor
market entry of one year, having a social scholarship could have been negatively correlated with the full year
of internships dummy but not correlated with wages. Furthermore, the choice of undertaking internships
during a full year or majoring in finance may depend on the conjuncture in the banking and insurance sector.
The value index (base 1 in December 2011) in June for French banking and insurance equities (AXA, Societe
Generale, Credit Agricole SA, BNP Paribas) could have been used as alternative instruments.
For IV analysis on wages, we perform an iterative choice of instruments. We begin with the student age
and the minor choice, then we add the year of admission, and at last a year of graduation higher than 2010.
All tests conclude to the consistency of instruments.
Results are reported in table 6 with different instruments, using 2SLS (GMM estimators give consistent
results). Columns 1, 2 and 3 are for the model with a dummy for the full year internship and IV analysis.
The full year of internships IV estimates are higher than the OLS ones, around 13-16% but standard errors
are often much higher too. Without taking into account the year of admission (column 1), the effect is not
significant. The column (2) with all instruments except year of graduation seems to be the best estimate
(no over identification, higher KP F-stat for weak identification). With IV analysis, the wage return of a full
year of internships appears higher than a real year of professional experience, but this result has only a local
interpretation. All instruments (except for the student age) and endogenous variables are indeed dummies
variables. The interpretation will be similar to a Local Average Treatment Effect (LATE), i.e. restricted to a
subpopulation. All first stages and the corresponding F-tests are reported in table 7. Results are consistent.
To be older of one year is associated with a decrease of taking-up a full year internship around 5-7%. Minoring
in mathematics is associated with an increase of majoring in finance around 45%. To be admitted in 2nd year
decreases the probability of a full year internship around 31-36% and the probability of majoring in finance
around 14%. A year of graduation higher than 2010 increases the take-up of internship of 31%.
15
5.4
Internships satisfaction at work outcomes
If internships enable students to learn about their preferences for a job, an industry, and to adapt in consequence their orientation, one would expect that optional internships would be associated with higher
satisfaction once at work. The surveys we use contain several questions on the degree of general satisfaction
at work, and satisfaction degrees regarding relations with colleagues (working environment), autonomy in
work, work conditions and salary level. Each satisfaction variable ranges from 1 (worst) to 5 (best). We note
that the response item 4 corresponds to a median response. What is important is that workers are more
satisfied than this median. We so consider a worker satisfied if he chooses the response item 4 or 5. We define
satisfaction for non monetary components with a score higher or equal than 4 for each component (working
environment, autonomy in work, and work conditions).
Tables 8 and 9 report OLS results of wages and non-pecuniary satisfaction on similar characteristics as
previously. Experience has a negative effect on satisfaction, which shows that expectations are not fulfilled.
Wages are highly correlated with wage satisfaction but not with non-pecuniary satisfaction. Having undertaken a full year of internships is positively related to non-pecuniary satisfaction and to wages satisfaction,
but the effects are not significant. Choosing a major in Finance is negatively correlated with wages satisfaction. An interesting point is that the non-pecuniary satisfaction is very difficult to explain. Unobserved
factors play an important role.
An absence of a link between internship and satisfaction may be possible if increasing the satisfaction at
work is not a selection criterion for doing an internship as the potential gains in non-monetary utility may
have other sources (prestige, altruism or social utility of a job). We may argue that satisfaction at work
may be badly anticipated by workers, which is consistent with the negative correlation between experience
and satisfaction. However, three critics arise regarding these results. First, tests lack of power. The nonsignificance may also be due to the sample size. Second, questions are measured with a Likert scale, i.e. an
ordered opinion on a subject, which may create a measurement issue. Aggregating the three components
of non-monetary satisfaction is valid only if these three components are highly correlated, which is only
partly the case with a correlation between 0.34-0.40. In order to check these assumptions, we performed a
disaggregated analysis by estimating a model for each of the three components (autonomy, working conditions and environment). We estimated probit models of a response item of 5 versus lower or equal than 4.
We constructed an alternative index of non-pecuniary satisfaction by summing the 3 satisfaction variables
regarding relations with colleague, autonomy in work, and work conditions (with several choices of points
for rare responses). In all cases, the effect of a full year of internship remains non-significant and low. The
third critic is that, as for wages, heterogeneous preferences can explain satisfaction at work, but also the
choice of doing a full year internship or majoring in finance. If students who did a full year internship or
choose a major in finance are those who have preferences for professional settings and lower preferences for
non monetary benefits, those students could engage in jobs with better wage and lower working conditions,
which may create an downward in the satisfaction equations. Hence, we performed an IV analysis similar to
the one conducted for wages (first stage regressions are close to the ones for wages). With this analysis, we
find indeed higher effects, but they remain not significant with high variability.
16
6
Conclusion
This article tests different predictions of internships effects using the endogenous selection process of an
optional one year internship. Perform a full year of internship is less valued by employers as a real year of
professional experience. Internships are perceived as a signal of ability by employers more than a return to
experience and a gain in human capital. This year of internship improves the ability to find a job faster.
The effects on job satisfaction remain ambiguous, positive but small and not significant. When a student
undertakes several internships in different areas, he/she is likely to refine his/her final major choices. These
questions are to be related to public policies. It opens the debate on compensation of internships, analyzing
returns to education for early career professionals. It also provides evidence on the time students need to
make effective major choices, and improve the social gain associated. Public policies implemented, which ban
unpaid internships or more than six months, appear consistent with our results.
Internships may reduce the uncertainty about the wage expectations for a given major and may reveal the
student preferences. But understanding the complete learning process of preferences associated with doing
a full year of internship (or shorter internships) is a tricky issue. The short and long internships may have
different effects. Furthermore, the role of non-market benefits such as working and learning in a high level
academic or professional environment or with international experts is difficult to assess quantitatively. At
last we do not model complete careers even if internships may generate better matching on the labor market
and better long-term opportunities.
17
7
Bibliography
Adda, Jerome, Christian Dustmann, Costas Meghir, and Jean-Marc Robin. (2010) “Career Progression and
Formal Versus On-the-Job Training.” Institute for Fiscal Studies Working Paper No. W10/13.
Alet, Elodie, and Liliane Bonnal. (2011) “Vocational Schooling and Educational Success: Comparing Apprenticeship to Full-time Vocational High School.” TSE Working Paper.
Altonji, Joseph G. (1993) “The Demand for and Return to Education when Education Outcomes are Uncertain.” Journal of Labor Economics, 11(1), 48-83.
Altonji, Joseph G. (1995) “The Effects of High School Curriculum on Education and Labor Market Outcomes.” The Journal of Human Resources, 30(3), 409-438.
Altonji, Joseph G., Erica Blom, and Costas Meghir. (2012) “Heterogeneity in Human Capital Investments:
High School Curriculum, College Major, and Careers.” NBER Working Paper No. 17985.
Arcidiacono, Peter. (2004) “Ability Sorting and the Returns to College Major.” Journal of Econometrics,
121(1-2), 343-375.
Arcidiacono, Peter, V. Joseph Hotz, and Songman Kang. (2012) “Modeling College Major Choices using
Elicited Measures of Expectations and Counterfactuals.” Journal of Econometrics, 166(1), 3-16.
Becker, Gary S. (1962) “Investment in Human Capital: a Theoretical Analysis.” Journal of Political Economics, 70(5), 9-49.
Beffy, Magali, Denis Fougère, and Arnaud Maurel. (2012) “Choosing the Field of Study in Postsecondary
Education: Do Expected Earnings Matter?” Review of Economics and Statistics, 94(1), 334-347.
Bishop, John H., and Ferran Mane. (2004) “The Impacts of Career-Technical Education on High School
Labor Market Success.” Economics of Education Review, 23(4), 381-402.
Björklund, Anders , and Christian Kjellström. (2002) “Estimating the Return to Investments in Education:
How Useful is the Standard Mincer Equation?” Economics of Education Review, 21, 195-210.
Blundell, Richard, Lorraine Dearden, and Barbara Sianesi. (2005) “Evaluating the Effect of Education on
Earnings: Models, Methods and Eesults from the National Child Development Survey.” Journal of the
Royal Statistical Society Series A, 168(3), 473-512.
Bonnal, Liliane, Sylvie Mendes, and Catherine Sofer. (2003) “Comparaison de l’Accès au Premier Emploi
des Apprentis et des Lycéens.” Annales d’Economie et de Statistique, 70, 32–52.
Dickson, Matt, and Colm Harmon. (2011) “Economic Returns to Education: What We Know, What We
Don’t Know, and Where We Are Going - Some Brief Pointers.” Economics of Education Review, 30,
1118-1122.
Fersterer, Josef, Jörn-Steffen Pischke, and Rudolf Winter-Ebmer. (2008) “Returns to Apprenticeship Training in Austria: Evidence from Failed Firms.” Scandinavian Journal of Economics, 110(4), 733-753.
18
Gaini, Mathilde, Aude Leduc, and Augustin Vicart. (2013) “School as a Shelter? School Leaving-Age and
the Business Cycle in France.” Annals of Economics and Statistics, 111/112, 251-270.
Gault, Jack, John Redington, and Tammy Schlager. (2000) “Undergraduate Business Internships and Career Success: Are They Related?” Journal of Marketing Education, 22(1), 45-53.
Gerken, Maike, Bart Rienties, Bas Giesbers, and Karen D. Könings. (2010) “Enhancing the Academic
Internship Learning Experience for Business Education - A Critical Review and Future Directions.” in
Learning at the Crossroads of Theory and Practice, Advances in Business Education and Training, vol.
4, Chap. 2, P. Van den Bossche et al. (eds.).
Haveman, Robert H., and Barbara L. Wolfe. (1984) “Schooling and Economic Well-Being: The Role of
Nonmarket Effects.” The Journal of Human Resources, 19(3), 377-407.
Heckman, James J., Lance J. Lochner, and Petra E. Todd. (2003) “Fifty Years of Mincer Earnings Regressions.” NBER Working Paper No. 9732.
Heckman, James J., and Sergio Urzúa. (2010) “Comparing IV with Structural Models: What Simple IV
can and cannot Identify.” Journal of Econometrics, 156, 27-37.
Mane, Ferran. (1999) “Trends in the Payoff to Academic and Occupation-Specific Skills: the Short and
Medium Run Returns to Academic and Vocational High School Courses for Non-College-Bound Students.” Economics of Education Review, 18(4), 417-437.
Narayanan, V. K., Paul M. Olk, and Cynthia V. Fukami. (2010) “Determinants of Internship Effectiveness:
An Exploratory Model.” Academy of Management Learning & Education, 9(1), 61-80.
Nunley, John M., Adam Pugh, Nicholas Romero, and Richard Alan Seals, Jr. (2014) “College Major,
Internship Experience, and Employment Opportunities: Estimates from a Résumé Audit.” Auburn
University Department of Economics Working Paper Series AUWP 2014-03.
Richmond, Tom. (2010) “Internships: To Pay or not to Pay?” Policy analysis and recommendations, The
Chartered Institute of Personnel and Development.
Rust, John. (1987) “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher.”
Econometrica, 55, 999-1033.
Saniter, Nils, and Thomas Siedler. (2014) “Door Opener or Waste of Time? The effects of Student Internships on Labor Market Outcomes.” IZA Discussion Paper No. 8141.
Solon, Gary, Steven J. Haider, and Jeffrey Wooldridge. (2013) “What Are We Weighting For? ” NBER
Working Paper No. 18859.
Stinebrickner, Todd R., and Ralph Stinebrickner. (2011) “Math or Science? Using Longitudinal Expectations Data to Examine the Process of Choosing a College Major.” NBERWorking Paper No. 16869.
Taylor, Susan M. (1988) “Effects of College Internships on Individual Participants.” Journal of Applied
Psychology, 73(3), 393-401.
19
Wolter, Stefan C., and Paul Ryan. (2011) “Apprenticeship.” in Handbook of the Economics of Education,
vol. 3, chap. 11, 521-576.
20
8
Tables and graphics
Note: figures from Analyse de l’orientation et des poursuites d’études des lycéens à partir de la procédure admission post-bac, Rapport
n° 2012-123, Inspection générale de l’éducation nationale, p.53
Figure 1: French higher education system
Note: full grey line for students admitted in first year, dash grey line for students admitted in second year, full black line for all
students.
Figure 2: Share of students undertaking a full year of internships.
21
Total
Full year
internship
No full year
internship
Mean Diff.
p-value
69.0
31.0
68.3
31.7
69.5
30.5
0.796
53.8
46.2
61.1
38.9
48.9
51.1
0.011
Ability of students (average inverse ranking)
First year of master
50.3
46.1
Minor choice
Mathematics 50.2
46.8
Economics
50.3
44.6
53.0
52.5
54.2
0.012
0.088
0.055
Minor and major choices (%)
Minor choice
(First year of master)
Mathematics
Economics
Major choice
(Second year of master) Finance
Other
Courses in Mathematics
Minor choice
Courses in Economics
Minor choice
# Obs.
Mathematics
Economics
52.2
56.9
41.5
49.9
54.3
40.5
53.7
58.7
42.2
0.111
0.099
0.680
Mathematics
Economics
52.7
51.9
54.4
50.8
52.1
47.9
53.9
51.7
58.9
0.168
0.889
0.007
452
180
272
Note: data from the administrative education software. For each first year of master (from 2004 to 2009), the best student obtained
the rank of 100, the worst gets 0. The global ranking is based on the average score during the year. Specific rankings are based on the
written individual evaluation. Courses included for the specific rankings have been defined using a principal component analysis.
Table 1: Minor, major choices (%) and schooling attainments
22
Total
Full year No full year
internship
internship
Labor market status 6 months after graduation (%)
Employment
63.2
73.4
55.3
Job search
6.0
2.4
8.7
Further studies
15.8
11.3
19.3
Ph.d
14.0
12.1
15.5
No activity
1.0
0.8
1.2
# Obs.
285
124
161
Duration of first job search (%)
Before graduating
Less than 2 months
Between 2 and 4 months
Between 4 and 6 months
More than 6 months
# Obs.
Total
Months after
graduation
3rd-year major field
6 months after
graduation
56.8
24.2
6.3
4.9
7.7
285
62.9
20.7
6.9
5.2
4.3
116
52.7
26.6
5.9
4.7
10.1
169
Average annual wages bonus included (in euros)
49.400
53.300
47.100
Mean Diff.
p-value
0.002
0.027
0.068
0.410
0.722
0.086
0.252
0.740
0.867
0.074
0.015
6 months
18 months
30 months
45.400
51.500
55.200
46.800
60.000
60.200
44.100
47.000
53.100
0.275
0.003
0.138
Finance
Other
# Obs.
47.300
42.200
444
48.700
43.100
167
45.900
41.500
277
0.435
0.433
72.5
79.8
88.1
59.9
63.3
327
77.9
80.2
89.3
65.7
64.9
131
68.9
79.6
87.2
56.1
62.2
196
0.113
0.917
0.612
0.137
0.679
Satisfaction at work (% of satisfied)
Total
Autonomy
Working conditions
Working environment
Non monetary components
Wages
# Obs.
Note: data from the survey "integration of graduates" from 2007 to 2012 (2007 excluded for satisfaction variables). For each satisfaction
variable (working conditions, working environment, autonomy in work, and the wages level), the scale ranges from 1 (worst) to 5 (best).
A worker is considered satisfied if the score is higher or equal than 4.
Table 2: Labor market outcomes after graduation
23
(1)
3rd year
Major
Finance
Probit
Full year intern.
(2)
3rd year
Major
Finance
Probit
0.140∗∗∗
(0.039)
Full year intern. Economics
(Ref. Full year intern. Finance)
Full year intern. Economics and Finance
2rd year minor: applied maths
(3)
3rd year
Major
Finance
Probit
0.049
(0.038)
(4)
3rd year
Major
Finance
Probit
0.052
(0.038)
-0.454∗∗∗
(0.049)
-0.435∗∗∗
(0.049)
-0.135∗
(0.069)
-0.128∗
(0.068)
0.292∗∗∗
(0.052)
0.523∗∗∗
(0.044)
0.525∗∗∗
(0.043)
0.340∗∗∗
(0.051)
0.002
(0.067)
0.034
(0.067)
0.110∗
(0.059)
2nd year inverse rank
Inverse rank - economics
(5)
3rd year
Major
Finance
Biprobit
0.047
(0.089)
(6)
Full year
Intern.
0.530∗∗∗
(0.044)
0.060
(0.045)
0.013
(0.070)
-0.184∗∗∗
(0.070)
Biprobit
-0.080
(0.080)
0.285∗∗∗
(0.078)
Inverse rank - mathematics
Graduating in 2008
(Ref. Grad. in 2007)
-0.100
(0.068)
-0.101
(0.067)
-0.049
(0.059)
-0.036
(0.059)
-0.101
(0.068)
0.067
(0.077)
Graduating in 2009
-0.089
(0.062)
-0.098
(0.061)
0.042
(0.053)
0.028
(0.053)
-0.091
(0.062)
0.053
(0.067)
Graduating in 2010
-0.111∗
(0.063)
-0.142∗∗
(0.063)
0.009
(0.055)
0.006
(0.054)
-0.121∗
(0.066)
0.257∗∗∗
(0.061)
Graduating in 2011
0.019
(0.061)
-0.030
(0.062)
0.041
(0.056)
0.035
(0.055)
0.004
(0.069)
0.360∗∗∗
(0.058)
Graduating in 2012
-0.113∗
(0.069)
-0.147∗∗
(0.068)
-0.047
(0.060)
-0.058
(0.060)
-0.122∗
(0.073)
0.268∗∗∗
(0.064)
Age in 2nd year
-0.072∗∗∗
(0.021)
Admitted in 2nd year
-0.288∗∗∗
(0.048)
10.654∗∗∗
(0.024)
452
0.269
0.537
Constant
Observations
Pseudo R2
Baseline probability
ρ̂
Note: Standard errors in parentheses.
∗
p < 0.10 ,
∗∗
p < 0.05 ,
10.554∗∗∗
(0.042)
452
0.288
0.537
∗∗∗
10.584∗∗∗
(0.027)
452
0.437
0.537
10.539∗∗∗
(0.046)
452
0.453
0.537
10.584∗∗∗
(0.036)
452
0.537
0.236
(0.194)
10.596∗∗∗
(0.046)
452
0.398
p < 0.01 . Columns 1, 2, 3 and 4 are the average marginal effects
of a probit model for a third year major in Finance. Columns 5 and 6 are the average marginal effects of a biprobit model for a third
year major in Finance and a full year of internship. Demographic characteristics (sex, place of birth) are used as control variables.
Table 3: Major determinants - probit and biprobit average marginal effects
24
(1)
Before
graduation
Probit
0.132∗∗
(0.059)
0.103∗
(0.062)
0.230∗∗
(0.097)
-0.212∗∗
(0.093)
-0.077
(0.090)
-0.130
(0.090)
-0.150
(0.092)
-0.272∗∗
(0.104)
Full year intern.
3rd year Major: Finance
2nd year inverse rank
Graduating in 2008
(Ref. Grad. in 2007)
Graduating in 2009
Graduating in 2010
Graduating in 2011
Graduating in 2012
Age in 2nd year
(2)
Before
graduation
IV 2SLS
0.202
(0.142)
-0.177
(0.117)
0.265∗∗
(0.105)
-0.214∗∗
(0.099)
-0.053
(0.089)
-0.158
(0.096)
-0.139
(0.103)
-0.322∗∗
(0.129)
2nd year minor: applied maths
Admitted in 2nd year
Constant
Observations
Pseudo R2 / Adjusted R2
Baseline probability
KP stat. for rank condition
285
0.067
0.569
(p-value)
Sargan-Hansen stat. for overid.
(p-value)
KP stat. for weak ident.
Note: Standard errors in parentheses.
∗
p < 0.10 ,
∗∗
p < 0.05 ,
0.614∗∗∗
(0.127)
285
0.569
54.898
0.000
0.119
0.730
26.644
∗∗∗
(3)
Full year
intern.
OLS
(4)
3rd year
Major: Finance
OLS
-0.184∗
(0.098)
0.089
(0.088)
0.043
(0.078)
0.217∗∗
(0.085)
0.399∗∗∗
(0.083)
0.420∗∗∗
(0.089)
-0.069∗∗∗
(0.023)
0.016
(0.062)
-0.335∗∗∗
(0.053)
2.051∗∗∗
(0.536)
285
0.216
0.407
27.08
0.000
0.075
(0.093)
-0.099
(0.094)
-0.030
(0.079)
-0.118
(0.085)
0.021
(0.085)
-0.056
(0.106)
0.013
(0.020)
0.569∗∗∗
(0.055)
-0.063
(0.054)
-0.054
(0.460)
285
0.321
0.607
36.42
0.000
(5)
Duration
Int. Reg.
-1.334∗
(0.684)
-0.510
(0.603)
-1.887∗
(1.041)
2.527∗∗∗
(0.950)
1.498
(0.967)
1.422
(0.900)
1.813∗
(0.944)
2.861∗∗∗
(1.096)
0.338
(0.636)
-1.012
(0.950)
285
.-
p < 0.01 . Column 1 is the average marginal effects of a probit
model for finding a job before graduation. Column 2 reports the 2SLS estimates with full year of internship and 3rd year Major in
Finance as endogenous variables and age in 2nd year, 2nd year minor-applied maths and admitted in 2nd year as instruments. Columns
3 and 4 report the 2SLS first-stages estimates for full year of internship and 3rd year Major in Finance. Column 5 is the estimates of
an interval regression of the duration of job search. KP for Kleibergen-Paap. Demographic characteristics (sex, place of birth) are used
as control variables.
Table 4: Duration of job search
25
Full year intern.
Experience
3rd year Major: Finance
2nd year inverse rank
Constant
Observations
Adjusted R2
KP stat. for rank condition
(p-value)
Sargan-Hansen stat. for overid.
(p-value)
KP stat. for weak ident.
Instruments
Age in 2nd year
2nd year minor: applied maths
Admitted in 2nd year
Year of graduation > 2010
Note: Standard errors in parentheses.
∗
p < 0.10 ,
∗∗
p < 0.05 ,
(1)
Log wage
IV 2SLS
0.151
(0.108)
0.109∗∗∗
(0.015)
0.080
(0.082)
0.092
(0.061)
10.515∗∗∗
(0.099)
444
0.354
20.642
(0.000)
.
.
10.283
(2)
Log wage
IV 2SLS
0.160∗∗
(0.076)
0.110∗∗∗
(0.015)
0.085
(0.072)
0.093
(0.059)
10.507∗∗∗
(0.063)
444
0.350
36.673
(0.000)
0.016
(0.900)
16.951
(3)
Log wage
IV 2LS
0.130∗∗
(0.053)
0.108∗∗∗
(0.015)
0.085
(0.071)
0.088
(0.056)
10.525∗∗∗
(0.057)
444
0.361
27.443
(0.000)
0.487
(0.784)
15.018
X
X
X
X
X
X
X
X
X
∗∗∗
p < 0.01 . Columns report the 2SLS estimates with full year of
internship and 3rd year Major in Finance as endogenous variables. Wages are deflated using price consumption index. The variancecovariance matrix is clustered with the student identifier. KP for Kleibergen-Paap. Job characteristics (working abroad, second job,
public sector, and fixed term contract) and demographic characteristics (sex, place of birth) are used as control variables.
Table 6: Wage determinants - IV 2SLS
26
27
10.654∗∗∗
(0.024)
444
0.348
76%
10.539∗∗∗
(0.046)
444
0.369
73%
10.584∗∗∗
(0.036)
444
0.369
43%
10.596∗∗∗
(0.046)
444
0.375
10.621∗∗∗
(0.057)
444
0.377
10.564∗∗∗
(0.041)
444
0.365
0.080
(0.053)
0.090∗∗∗
(0.030)
Table 5: Wages determinants and interaction effects - OLS and LAD
Note: Standard errors in parentheses. ∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01 . Columns 1, 2, and 4 report the OLS estimates without interraction effects, columns 5, 6, 7, and 8 the OLS estimates with
interaction effects, column 3 the LAD estimates. Wages are deflated using price consumption index. The variance-covariance matrix is clustered with the student identifier (except for the LAD estimator). Job
characteristics (working abroad, second job, public sector, and fixed term contract) and demographic characteristics (sex, place of birth) are used as control variables.
Observations
Adjusted R2
Full year intern./Experience
Constant
0.249∗∗
(0.125)
-0.130∗
(0.077)
0.072
(0.070)
Inverse rank - mathematics x Full year intern.
0.196∗
(0.102)
-0.004
(0.067)
0.104∗∗∗
(0.032)
0.052
(0.141)
10.584∗∗∗
(0.027)
444
0.074
(0.054)
0.094∗∗∗
(0.031)
Inverse rank - economics x Full year intern.
10.554∗∗∗
(0.042)
444
0.368
73%
-0.028
(0.058)
Inverse rank - mathematics
2nd year inverse rank x Full year intern.
0.127∗∗
(0.062)
0.052∗
(0.031)
Inverse rank - economics
0.078
(0.052)
0.015
(0.032)
Full year intern. x Experience
2nd year inverse rank
0.057∗∗
(0.028)
0.051
(0.039)
Finance x Experience
0.096∗∗∗
(0.031)
-0.003
(0.063)
0.069∗∗∗
(0.021)
Full year intern. x Finance
0.089∗∗∗
(0.030)
0.290∗∗∗
(0.060)
Full year intern. - 2 years experience
3rd year Major: Finance
0.201∗∗∗
(0.032)
No full year intern. - 2 years experience
0.105∗∗∗
(0.015)
(8)
Log wage
OLS
0.185∗∗∗
(0.040)
0.108∗∗∗
(0.015)
(7)
Log wage
OLS
-0.080
(0.079)
Full year intern. - 1 year experience
0.066∗∗∗
(0.019)
(6)
Log wage
OLS
-0.017
(0.057)
0.079∗∗∗
(0.027)
0.103∗∗∗
(0.015)
(5)
Log wage
OLS
0.067
(0.049)
No full year intern. - 1 year experience
0.090∗∗∗
(0.012)
(4)
Log wage
OLS
0.075∗∗
(0.029)
0.053
(0.034)
0.106∗∗∗
(0.015)
0.105∗∗∗
(0.015)
(3)
Log wage
LAD
0.039∗∗
(0.018)
Full year intern. - No experience
Experience
Full year intern.
(2)
Log wage
OLS
0.077∗∗∗
(0.030)
(1)
Log wage
OLS
0.080∗∗∗
(0.029)
Age in 2nd year
2nd year minor: applied maths
Admitted in 2nd year
(1)
Full year
intern.
(2)
Full year
intern.
(3)
Full year
intern.
OLS
-0.119∗∗∗
(0.023)
-0.032
(0.078)
OLS
-0.069∗∗∗
(0.024)
-0.035
(0.069)
-0.308∗∗∗
(0.063)
3.392∗∗∗
(0.536)
444
0.147
13.98
2.344∗∗∗
(0.545)
444
0.222
19.63
OLS
-0.053∗∗
(0.023)
-0.039
(0.065)
-0.363∗∗∗
(0.057)
0.314∗∗∗
(0.055)
1.834∗∗∗
(0.552)
444
0.313
33.78
Year of graduation > 2010
Constant
Observations
Adjusted R2
F-Stat
Note: Standard errors in parentheses data,
∗
p < 0.10 ,
∗∗
p < 0.05 ,
∗∗∗
(4)
3rd year
Major
Finance
OLS
0.014
(0.018)
0.451∗∗∗
(0.069)
(5)
3rd year
Major
Finance
OLS
0.037∗
(0.020)
0.449∗∗∗
(0.067)
-0.142∗∗
(0.059)
-0.033
(0.419)
444
0.430
22.63
-0.518
(0.451)
444
0.444
19.57
(6)
3rd year
Major
Finance
OLS
0.037∗
(0.020)
0.449∗∗∗
(0.067)
-0.142∗∗
(0.059)
0.001
(0.052)
-0.519
(0.462)
444
0.443
14.95
p < 0.01 . Columns report the 2SLS first-stages estimates
for full year of internship and 3rd year Major in Finance. Wages are deflated using price consumption index. The variance-covariance
matrix is clustered with the student identifier.
Table 7: Wage determinants - IV 2SLS first-stages
28
Full year intern.
3rd year Major: Finance
Experience
2nd year inverse rank
Satisfaction
Wage
Satisfaction
Wage
Satisfaction
Wage
Satisfaction
Wage
Full year
intern.
Probit
0.041
(0.062)
-0.145∗∗
(0.073)
-0.033
(0.034)
0.183∗
(0.104)
Probit
0.003
(0.060)
-0.174∗∗
(0.070)
-0.093∗∗
(0.036)
0.124
(0.100)
0.568∗∗∗
(0.169)
IV 2SLS
0.142
(0.101)
-0.421∗∗∗
(0.132)
-0.028
(0.037)
0.222∗∗
(0.113)
IV 2SLS
0.078
(0.107)
-0.452∗∗∗
(0.125)
-0.081∗∗
(0.040)
0.108
(0.188)
0.501∗∗∗
(0.135)
OLS
3rd year
Major
Finance
OLS
0.021
(0.026)
-0.068
(0.121)
-0.015
(0.022)
0.191∗
(0.098)
-0.054∗∗
(0.025)
0.019
(0.069)
-0.428∗∗∗
(0.064)
0.335∗∗∗
(0.061)
1.786∗∗∗
(0.585)
327
0.364
0.376
36.56
(0.000)
0.026
(0.021)
0.515∗∗∗
(0.072)
-0.144∗∗
(0.061)
0.028
(0.054)
-0.347
(0.493)
327
0.512
0.597
17.23
(0.000)
Log wage
Age in 2nd year
2nd year minor: applied maths
Admitted in 2nd year
Year of graduation > 2010
Constant
Observations
Pseudo R2 / Adjusted R2
Baseline probability
KP stat. for rank condition
(p-value)
Sargan-Hansen stat for overid.
(p-value)
KP stat. for weak ident.
327
0.079
0.633
327
0.121
0.633
0.774∗∗∗
(0.121)
327
0.018
0.633
23.134
(0.000)
0.743
(0.690)
13.142
-4.506∗∗∗
(1.415)
327
0.059
0.633
22.846
(0.000)
0.945
(0.623)
12.757
Note: Standard errors in parentheses. ∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01 . Columns 1 and 2 report the average
marginal effects of a probit model for beeing satisfied with wages (i.e. a score higher than 4). Columns 3 and 4 report
the 2SLS estimates with full year of internship and 3rd year Major in Finance as endogenous variables and age in 2nd
year, 2nd year minor-applied maths, admitted in 2nd year, and year of graduation as instruments. Columns 5 and 6
report the 2SLS first-stages estimates (without wage) for full year of internship and 3rd year Major in Finance. Wages
are deflated using price consumption index. The variance-covariance matrix is clustered with the student identifier.
KP for Kleibergen-Paap. Job characteristics (working abroad, second job, public sector, and fixed term contract) and
demographic characteristics (sex, place of birth) are used as control variables.
Table 8: Job satisfaction - Wage
29
Full year intern.
3rd year Major: Finance
Experience
2nd year inverse rank
Satisfaction
Non mon.
Satisfaction
Non mon.
Satisfaction
Non mon.
Satisfaction
Non mon.
Full year
intern.
Probit
0.084
(0.064)
-0.007
(0.080)
-0.078∗∗
(0.035)
0.016
(0.114)
Probit
0.071
(0.064)
-0.015
(0.083)
-0.094∗∗
(0.037)
-0.001
(0.115)
0.155
(0.161)
IV 2SLS
0.112
(0.113)
-0.052
(0.172)
-0.075∗∗
(0.036)
0.026
(0.116)
IV 2SLS
0.092
(0.115)
-0.062
(0.176)
-0.091∗∗
(0.038)
0.011
(0.116)
0.156
(0.163)
OLS
3rd year
Major
Finance
OLS
0.631∗∗∗
(0.147)
327
0.021
0.599
23.134
(0.000)
2.297
(0.317)
13.142
-1.016
(1.694)
327
0.023
0.599
22.846
(0.000)
2.471
(0.291)
12.757
0.002
(0.028)
-0.086
(0.123)
0.184
(0.112)
-0.052∗∗
(0.024)
0.012
(0.068)
-0.415∗∗∗
(0.064)
0.330∗∗∗
(0.061)
-0.206
(1.326)
327
0.369
0.376
33.37
(0.000)
-0.028
(0.024)
0.179∗
(0.101)
0.121
(0.100)
0.027
(0.021)
0.511∗∗∗
(0.074)
-0.136∗∗
(0.062)
0.024
(0.054)
-1.654
(1.239)
327
0.514
0.597
15.57
(0.000)
Log wage
Age in 2nd year
2nd year minor: applied maths
Admitted in 2nd year
Year of graduation > 2010
Constant
Observations
Pseudo R2 / Adjusted R2
Baseline probability
KP stat. for rank condition
(p-value)
Sargan-Hansen stat for overid.
(p-value)
KP stat. for weak ident.
327
0.041
0.599
327
0.045
0.599
Note: Standard errors in parentheses. ∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01 . Columns 1 and 2 report the
average marginal effects of a probit model for beeing satisfied with all non monetary components (i.e. a score higher
than 4 for autonomy in work, working conditions and environment). Columns 3 and 4 report the 2SLS estimates
with full year of internship and 3rd year Major in Finance as endogenous variables and age in 2nd year, 2nd year
minor-applied maths, admitted in 2nd year, and year of graduation as instruments. Columns 5 and 6 report the 2SLS
first-stages estimates (with wage) for full year of internship and 3rd year Major in Finance. Wages are deflated using
price consumption index. The variance-covariance matrix is clustered with the student identifier. KP for KleibergenPaap. Job characteristics (working abroad, second job, public sector, and fixed term contract) and demographic
characteristics (sex, place of birth) are used as control variables.
Table 9: Job satisfaction - Non monetary components
30